Published on June 17, 2026 at 03:50 CEST (UTC+2)
GrapheneOS has been ported to Android 17 (384 points by Cider9986)
GrapheneOS has been ported to Android 17 – The security-focused Android ROM GrapheneOS announces a full port to Android 17, with official releases imminent. The team has already tested the port on multiple Pixel devices (6a, 7, 8, 10a, etc.) and is pushing code to public repositories. Community discussion clarifies that while Google may drop support for older Pixels like the 6/6 Pro, GrapheneOS aims to maintain broader device compatibility.
Running local models is good now (1018 points by jfb)
Running local models is good now – Vicki Boykis argues that local LLMs have finally become practical for real development work, especially with models like GPT‑OSS and Google’s Gemma 4. On a 2022 M2 Mac, she achieves roughly 75% of frontier model accuracy/speed for agentic coding tasks. She now uses local models as a fast, personalized search engine for coding questions and has successfully refactored entire scripts locally without turning to cloud APIs.
Humiliating IIS servers for fun and jail time (86 points by denysvitali)
Humiliating IIS servers for fun and jail time – A bug bounty hunter details systematic techniques for finding and exploiting misconfigured IIS web servers. Methods include Shodan, Google dorking, tilde enumeration, web.config exposure, and path traversal. The author also describes using LLMs and BigQuery to resolve shortnames, and recommends automating reconnaissance with nuclei templates.
SpaceX to buy Cursor for $60B (881 points by itsmarcelg)
SpaceX to buy Cursor for $60B – Reuters reports that SpaceX is acquiring Anysphere, the company behind the AI code assistant Cursor, for $60 billion. The deal highlights the extreme valuation placed on AI‑powered development tools. It suggests SpaceX intends to integrate advanced AI copilot capabilities into its engineering and operations workflows.
Wolfram Language and Mathematica Version 15, AI Assistant, Symbolic Music, More (62 points by alok-g)
Wolfram Language and Mathematica Version 15, AI Assistant, Symbolic Music, More – Stephen Wolfram announces Version 15, featuring an integrated AI assistant within every notebook, symbolic music capabilities, and extensive new functionality. The release also includes time‑series improvements, categorical data support, a new ModelFit superfunction, and structured package formats. The AI assistant aims to make Wolfram Language accessible directly from natural language interfaces.
TIL: You can make HTTP requests without curl using Bash /dev/TCP (265 points by mrshu)
TIL: You can make HTTP requests without curl using Bash /dev/TCP – A practical tip demonstrates using Bash’s built‑in /dev/tcp pseudo‑device to make raw HTTP requests from minimal containers lacking curl or wget. The method opens a TCP socket, sends a manually crafted HTTP request (including headers like Authorization), and reads the response. The trick is useful for debugging and health‑check scenarios in stripped‑down Docker images.
Calvin and Hobbes and the price of integrity (282 points by pseudolus)
Calvin and Hobbes and the price of integrity – An in‑depth essay explores how Bill Watterson maintained uncompromising moral integrity while creating Calvin and Hobbes, including his refusal to license merchandise despite enormous financial pressure. It traces his artistic journey from college to syndication, highlighting his philosophy of treating the comic strip as art rather than commerce. The piece reflects on the personal and professional costs of sticking to one’s principles in a commercial industry.
Mechanical Watch (2022) (625 points by razin)
Mechanical Watch (2022) – An interactive, visual guide explains the inner workings of a mechanical watch movement, from the mainspring to the escapement. The author uses 3D animations and color‑coded components to demystify the jargon‑heavy world of horology. The article emphasizes that mechanical watches operate without batteries or electronics, relying purely on mechanical ingenuity.
Stop Using JWTs (248 points by dzonga)
Stop Using JWTs – A developer argues that JSON Web Tokens (JWTs) are frequently misused for session management, leading to problems with revocation, payload size, and complexity. The gist advocates for simpler opaque session tokens stored in cookies, which are easier to invalidate and less prone to security pitfalls. It calls for a rethink of authentication patterns in modern web applications.
GPT‑NL: a sovereign language model for the Netherlands (135 points by root-parent)
GPT‑NL: a sovereign language model for the Netherlands – TNO, together with SURF and the Netherlands Forensic Institute, is building an independent Dutch language model called GPT‑NL. The project prioritizes transparency, privacy, copyright compliance, and public values over commercial control. It aims to strengthen Europe’s digital autonomy by providing a trusted, governable alternative to models from large US tech companies.
Local models have reached a practical tipping point – The article “Running local models is good now” shows that with the latest models (Gemma 4, GPT‑OSS), locally‑run LLMs can handle agentic coding at ~75% of frontier performance. This is a major shift: developers can now keep sensitive code entirely on‑device, reduce API costs, and work offline. Implication: Expect rapid adoption of local models for productivity tools, with hardware upgrades (high‑RAM Macs, etc.) becoming a new standard for developers.
AI code assistants are becoming strategic acquisitions – SpaceX’s $60B purchase of Cursor (Anysphere) signals that AI‑powered coding is viewed as a core competitive advantage, even for non‑software companies. The deal validates the thesis that copilot tools will dramatically accelerate engineering workflows. Implication: More large enterprises will acquire or build custom AI coding assistants; the market for such tools will see consolidation and increased valuation pressure.
Sovereign AI models address geopolitical and regulatory needs – GPT‑NL in the Netherlands exemplifies a growing trend: nations and regions building their own LLMs to ensure data governance, privacy, and alignment with local laws (e.g., GDPR). This counters reliance on US‑ or China‑based models. Implication: Expect more sovereign AI initiatives (e.g., for smaller languages, public sector use) to emerge, potentially fragmenting the LLM landscape but improving trust and compliance.
AI is being embedded into traditional computational platforms – Wolfram’s Version 15 integrates an AI assistant directly into Mathematica notebooks, blurring the line between computational engines and conversational interfaces. This move mirrors what we see in IDEs (Copilot) and office suites (Microsoft Copilot). Implication: Every major tool with a scripting or query interface will soon have an embedded AI layer, changing how users interact with complex software—shifting from command‑based to intent‑based workflows.
LLMs are increasingly used in security research and exploitation – The IIS bug‑bounty article explicitly uses LLMs to resolve shortnames via GitHub dorks and BigQuery, demonstrating how AI accelerates vulnerability discovery. Attackers and defenders alike are integrating LLMs into their toolchains. Implication: Security teams must adopt AI‑augmented automation to keep up; the cat‑and‑mouse game between exploit generation and detection will intensify, and LLM‑specific security audits (e.g., prompt injection in security tools) will become essential.
Authentication debates are resurfacing in an AI‑heavy world – The “Stop Using JWTs” gist, while not directly about AI, reflects a growing need for simpler, more revocable authentication patterns that work well with AI agents and automated API calls. As AI agents make more server‑side requests, token management (size, expiry, revocation) becomes critical. Implication: Expect a push toward session‑based or cookie‑based authentication for machine‑to‑machine interactions, and possibly new standards tailored for AI‑driven microservices.
“Vibe metrics” are replacing benchmark‑only model evaluation – Vicki Boykis’s personal metric—“Do I have to double‑check against an API model?”—represents a broader shift toward user‑centric, practical quality assessment. The community is moving beyond static benchmarks (MMLU, HumanEval) to subjective utility measures. Implication: Model developers and vendors will increasingly highlight real‑world use cases over academic scores; product decisions may be driven by anecdotal “feels good enough” data, which could accelerate model adoption but also risk hidden failure modes.
Analysis generated by deepseek-reasoner